model-writtenrun 6208 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut announced a new psql query-cancellation regression test after an earlier cancellation regression had been reverted, and Daniel Gustafsson promptly reported a buildfarm failure on sidewinder #1 #2.
- Fabien COELHO questioned whether the SKIP block should tolerate timing failures, Daniel explained that it only skipped Windows, and Tom Lane argued that suppressing failures would defeat the test and recommended replacing SIGALRM with bounded PID-file polling #3 #4 #5.
- Daniel suggested interactive_psql, while Tom investigated possible inherited-alarm behavior on NetBSD and subsequently traced wrasse's failure to an unexpected SIGTERM after SIGINT, reinforcing the need to remove the signal handler #6 #7 #8.
- Daniel supplied an interactive_psql implementation that still assumed the sleep query had started after a timer expired; Peter reported failures, and Tom reproduced a missing cancellation error on wrasse and objected to the remaining timing assumptions #9 #10 #12.
- Peter replaced that approach with asynchronous IPC::Run execution, waiting separately for the PID file and for the sleep query to appear in pg_stat_activity; Tom confirmed it passed on wrasse, and Daniel endorsed the design #13 #14 #15.
- Tom requested checking that the PID file contained a complete newline-terminated value, and Fabien supplied a revised patch that read and validated the file inside the polling loop #14 #16.
- Peter committed Fabien's revision with a small restructuring of the PID wait loop as 43d4dd8797 and subsequently reported that the failures appeared fixed #16 #18.
- Michael Paquier later reported another cancellation-test failure on conchuela, and Daniel identified an empty error output as suggesting another timing issue; the thread ends with that failure unresolved #19 #20.